The University of Arizona
Please note that this event has ended!

Connectionist Temporal Classification

Multi-lingual Optical Character Recognition Seminar

Connectionist Temporal Classification
Series: Multi-lingual Optical Character Recognition Seminar
Location: ENR2 S395
Presenter: Marek Rychlik

CTC (Connectionist Temporal Classification)
          is a probability model for segmentation of outputs
          of recurrent neural networks into meaningful
          chunks. The technique is used for handwriting and
          script language segmentation, and for natural
          speech. In Deep Learning, CTC is simply a kind of
          rather sophisticated loss function. I will discuss
          the paper of Alex Graves (currently at Deep Mind)
          who introduced this method about 10 years ago to
          handwriting recognition and OCR.